Agent Based Modeling of Seismic Time Series Data

Abstract

Agent Based Modeling of Seismic Time Series Data Earthquake activities are taking place all around the globe. Latest disastrous series of earthquakes in Nepal in April and May 2015, earthquakes in Pakistan during October 2005 and 2015, and various other large events serve as ruling reasons to investigate the causes of these dangers. After every such disaster, a new chapter of research is opened. Predicting such events has always been a complicated task as it involves many underlying systems. The prediction itself does not reduce the chances of earthquakes but yields alarms before the danger. Prediction alarms can save precious lives and unfurl sense of preparation among people. This work is an endeavor in this regard. With the novelty of approaches using Multi-Agent Systems (MAS) with other models; like, Swarm Intelligence and Poisson distribution, future earthquakes are analyzed. Recent research shows uses of Intelligent Agents for the prediction of ecosystem management, forecasting and scheduling of aero engine overhaul and analysis and prediction of natural resource management. Several authors applied evolutionary techniques like Particle Swarm Optimization, Neural Networks and Genetic Algorithms for earthquake prediction and analysis. In this thesis, Multi-Agent based Prediction Model (MAP) is developed with the statistically inferred rules and reasoning. MAP does not only help to predict the nature of future events but also has the capability of real world phenomenon modeling. Scientific explanations for earthquakes, which needed expensive infrastructure and equipment before, can be visualized as Intelligent Agents (IA) with associated behavior and learning capability. This work elaborates various uses of Multi-Agents along with time series data of earthquake events. Algorithms used in the conjugation of Multi-Agents are Bare Bone Particle Swarm Optimization (BPSO), Poisson distribution, and statistically inferred rules. Several new parameters are introduced to work with the Multi-Agents. Shape and distance parameters are used with latitude and longitude based MAP resulting 93.1% accurate prediction. Data from United States Geological Survey (USGS), Advanced National Seismic System (ANSS) have been used for analyses. The behavior of each agent is designed based upon statistically inferred results. The relationship between statistical inference and visually scaled parameters is drawn. The prediction system resembles real world scenarios. Higher and medium intensity earthquakes are predicted. In another technique, Enhanced version of BPSO is proposed as EBPSO to work with MAS. Different parameters tested though EBPSO are depth, magnitude with respect to time, magnitude sorted in the order of latitude and longitude and day difference between consecutive earthquakes. Results obtained from EBPSO are later analyzed using IA. High-risk and Low-risk areas are identified using this model. Using the Gaussian distribution of BPSO with standard error adjustment, 91-98% prediction accuracy is achieved for different parameters. Another technique, Poisson distribution of magnitude on latitude and longitude is calculated. Later, agents are designed to work on the results of Poisson distribution to sense the neighboring areas. This technique also yielded 93% optimum results on high x intensity. Then, correlation between several formal and novel parameters is investigated to identify dangerous areas. In short, MAP outperformed the existing techniques to draw a conceptual relationship between different parameters. Experiments show that EBPSO has high accuracy rate as compared to others similar techniques. Poisson distribution used with IA gives the likeliness of high and medium intensities.